fpgamlirfanboy

6 ms·

31.
by fpgamlirfanboy 3y ago
I promise I'm not trying to rain on your parade, and I blame the academics for this, but: 1. NP-complete problems are (today) global search problems (3-SAT algos search for a satisfying assignment...); 2. Combinatorial search spaces
32.
by fpgamlirfanboy 3y ago
It's amazing how heavily provided hn is. I have a response here that's been deleted that is like 15 words, including a link to source that corroborates my claim but that response contains a transcribed emoji and so it's been
33.
by fpgamlirfanboy 3y ago
> zero cost abstractions exist So does perpetual motion :shrug: but my point is Triton is not an abstraction in the least. Source: 1) I spent 6 months investigating targeting other backends 2) Phil himself said he doesn't care to su
34.
by fpgamlirfanboy 3y ago
> allows a level of abstraction that doesn't affect performance even a little bit The second part of this sentence is true because the first part is false.
35.
by fpgamlirfanboy 3y ago
> needing to look up the definition Who has ever needed to look up the definition of something like "impactful" or "observability"...?
36.
by fpgamlirfanboy 3y ago
I hate all of these goofy neologisms corporate culture comes up with ("impactful") that are just abuses of syntax and have no reason to exist because there's already a word that means the same thing. In this case the word
37.
by fpgamlirfanboy 3y ago
> LLVM/GCC target fixed ISA devices. Even GPUs can be represented in this paradigm of having a core set of instructions and registers for SIMD/SMT. What FPGA or random ASIC could LLVM front-end? How do you go from LLVM-IR to so
38.
by fpgamlirfanboy 3y ago
ghc doesn't target C since 7.0. the other two i couldn't be bothered to care even less about. when you can point to a production grade """compiler""" that targets C (instead of emitting ir or asm dire
39.
by fpgamlirfanboy 3y ago
> Until the "industry" values hardware industries don't value things, markets value things. RTL work doesn't pay well because the products don't sell into markets that scale like software. with the exception of N
40.
by fpgamlirfanboy 3y ago
> It sucks, but it’s everywhere and tied into everything. meh lots of places aren't actually programming system verilog - they're using python/perl to generate it. that's not the same thing.
41.
by fpgamlirfanboy 3y ago
> that doesn't make them perl scripts! but it doesn't make them compilers either :shrug:
42.
by fpgamlirfanboy 3y ago
> The only scalable/vendor/device-neutral way to do that is to back-end your "new" HDL with Verilog/SystemVerilog/VHDL i mean like the guy below says, but in the opposite tone, that's like saying buildi
43.
by fpgamlirfanboy 3y ago
i'll hop on the dismissive bandwagon but not for the same reason as everyone else: not a single one of these HDLs ever (ever) amount to more than perl scripts. why? how? how could that possibly be? this one is written in rust...? becau
44.
by fpgamlirfanboy 3y ago
> are there any HLS systems that you think are good enough to implement most of the techniques we see in the AI research papers? Especially those reducing the cost of pre-training or increasing the context of models. you're talking
45.
by fpgamlirfanboy 3y ago
i work in both of these areas (compilers [contrib on iree/mlir] and RTL); they're not even remotely doing the same thing. iree is a runtime environment (Intermediate Representation Execution Environment) - it's not even a com